Automated CVE Monitoring with LLM-Based Impact Analysis and Real-Time Telegram Alerting: A System Design and Proof-of-Concept

Abstract

The rapid increase in publicly disclosed cybersecurity vulnerabilities presents significant challenges for security teams due to the time-consuming manual triage of Common Vulnerabilities and Exposures (CVE) data. This work addresses the need for an automated, end-to-end CVE monitoring system capable of continuous data acquisition, intelligent impact analysis, and real-time notification. The proposed framework integrates an NVD crawler retrieving multi-version CVSS metrics, a DeepSeek Large Language Model (LLM) module for contextual impact analysis, and a Telegram bot for structured real-time alerts. Evaluation of the proof-of-concept confirmed robust operation, achieving a 100% notification delivery success rate with an average end-to-end processing latency of 5.7 seconds. Structured evaluation of the LLM integration across 200 CVE samples yielded a mean composite quality score of 2.74 out of 3.0 and an 89.5% operational pass rate, demonstrating exceptional capability on well-defined weaknesses. The results indicate that LLM-augmented monitoring significantly reduces manual triage effort while maintaining real-time operational awareness, establishing a practical foundation for scalable autonomous vulnerability management workflows.